Remove channels_last requirement for pool ops via explicit permutes (#21796)

[mcremon-meta]

Summary:

Changes the cortex_m pool ops (avg_pool2d, max_pool2d) from the
"logical-NCHW-shape + NHWC-physical (channels_last dim_order)" contract to a
"logical-NHWC-shape, plain contiguous" contract, bridged to the surrounding NCHW
graph by explicit aten.permute_copy nodes. This lets the pool ops run when the
graph is contiguous (e.g. _skip_dim_order=True or channels_last not propagated),
which previously failed at runtime with:
cortex_m_ops_common.h: quantized_avg_pool2d_out: tensors must use channels_last
dimension order

The CMSIS pool kernels already treat the data pointer as physically NHWC; the
physical bytes are identical in both contracts, so the kernel change is only how
dims are read (NHWC positions) plus dropping the channels_last check.

Changes (mirrored fbcode + xplat, DirSync-identical):

• ops/cortex_m_ops_common.h (prepare_cmsis_pool2d_config): read dims from NHWC
  positions (size(1)=H, size(2)=W, size(3)=C), compare batch/channels at
  size(0)/size(3), and drop the require_channels_last parameter/check.
• ops/op_quantized_avg_pool2d.cpp: update call site; NHWC-consistent.
• ops/operators.py: pool fake-metas use NHWC-shaped I/O; the avg_pool2d and
  max_pool2d reference impls permute NHWC<->NCHW around F.*_pool2d and return
  NHWC-shaped output (consistent with the NHWC kernel contract).
• passes/aten_to_cortex_m_pass.py: pool replacements insert explicit
  permute_copy (NCHW->NHWC before, NHWC->NCHW after); pad reworked for NHWC;
  qparams preserved.
• passes/scratch_buffer_sizes.py: pool scratch sizing uses NHWC-shaped I/O.

Differential Revision: D113943678

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/21796

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 2 New Failures

As of commit 75262bb with merge base 43f89fb ():

NEW FAILURES - The following jobs have failed:

• pull / test-arduino-library / run / arduino-library (gh)
  RuntimeError: Command docker exec -t b003dc10bccd2e57426a16c6dacfa4bd1b51d190ed649f44ab1812dbc2bfd242 /exec failed with exit code 1
• pull / test-mcu-cortex-m-backend / linux-job (gh)
  RuntimeError: Command docker exec -t 8f44d40fd3a06eccea63365f8b82fc3fda54f8f1a6936d9f6c717c62f2a03fb7 /exec failed with exit code 1

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[meta-codesync]

@mcremon-meta has exported this pull request. If you are a Meta employee, you can view the originating Diff in D113943678.

[github-actions]

This PR needs a release notes: label

If your change should be included in the release notes (i.e. would users of this library care about this change?), please use a label starting with release notes:. This helps us keep track and include your important work in the next release notes.

To add a label, you can comment to pytorchbot, for example
@pytorchbot label "release notes: none"

For more information, see
https://github.com/pytorch/pytorch/wiki/PyTorch-AutoLabel-Bot#why-categorize-for-release-notes-and-how-does-it-work.

[mcremon-meta]

@AdrianLundell do you mind taking a look when you're back and get a chance? :)